CN110570487B - Undersampling model generation method, image reconstruction method, device and computer equipment - Google Patents
Undersampling model generation method, image reconstruction method, device and computer equipment Download PDFInfo
- Publication number
- CN110570487B CN110570487B CN201910789671.3A CN201910789671A CN110570487B CN 110570487 B CN110570487 B CN 110570487B CN 201910789671 A CN201910789671 A CN 201910789671A CN 110570487 B CN110570487 B CN 110570487B
- Authority
- CN
- China
- Prior art keywords
- model
- undersampled
- image
- reconstructed image
- updated
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T11/00—2D [Two Dimensional] image generation
- G06T11/003—Reconstruction from projections, e.g. tomography
- G06T11/008—Specific post-processing after tomographic reconstruction, e.g. voxelisation, metal artifact correction
Abstract
The application relates to an undersampled model generation method, an image reconstruction device, computer equipment and a storage medium. The method comprises the following steps: obtaining an undersampled image according to the undersampled model to be updated; reconstructing the undersampled image through a preset neural network to obtain a reconstructed image; when the reconstructed image does not meet the preset requirement, acquiring a coding line with the minimum difference value; updating the under-sampling model to be updated according to the coding line to generate a new under-sampling model; and taking the new undersampled model as an undersampled model to be updated, returning to the step of obtaining an undersampled image according to the undersampled model to be updated until the reconstructed image meets the preset requirement, and taking the undersampled model corresponding to the reconstructed image meeting the requirement as the finally generated undersampled model. By adopting the method, the matching rate of the undersampling model and the neural network can be improved.
Description
Technical Field
The present application relates to the field of medical technology, and in particular, to an under-sampling model generation method, an image reconstruction method, an apparatus, a computer device, and a storage medium.
Background
With the development of the medical technical field, various medical imaging technologies are developed, and clear medical images can be obtained by performing image reconstruction on data scanned and acquired by imaging equipment, so that a user can conveniently predict a focus. Especially, magnetic Resonance Imaging (MRI) is an irreplaceable member of modern medical imaging technology because it has no ionizing radiation damage and possesses various tissue contrasts. The process of magnetic resonance image acquisition is typically to spatially encode the entire K-space, resulting in image acquisition times that are often long. However, if the acquisition time is long, a series of influences are caused, for example, the motion artifact influences the image quality. Therefore, how to increase the sampling speed of the magnetic resonance scan is always a hot spot in the field.
However, since the central low-frequency part of the K-space contains the main structure and contrast, and the high-frequency part contains the physical laws of detail, an undersampled model is usually adopted for data acquisition of undersampling the high-frequency part. However, with the development of artificial intelligence technology, image reconstruction of acquired magnetic resonance data through a neural network is a new magnetic resonance reconstruction method. However, neural networks are data-driven technologies, so that there is no particular need for under-sampled models, resulting in the inability to select an under-sampled model that matches the neural network.
Disclosure of Invention
In view of the above, it is desirable to provide an undersampled model generation method, an image reconstruction method, an apparatus, a computer device, and a storage medium that are capable of reducing reconstruction artifacts and are most suitable for a neural network.
A method of undersampled model generation, the method comprising:
obtaining an undersampled image according to the undersampled model to be updated;
reconstructing the undersampled image through a preset neural network to obtain a reconstructed image;
when the reconstructed image does not meet the preset requirement, acquiring a coding line with the minimum difference value;
generating a new undersampling model according to the coding line with the minimum difference value;
and taking the new undersampled model as an undersampled model to be updated, returning to the step of obtaining an undersampled image according to the undersampled model to be updated until the reconstructed image meets the preset requirement, and taking the undersampled model corresponding to the reconstructed image meeting the requirement as the finally generated undersampled model.
A method of image reconstruction, the method comprising:
obtaining an image to be reconstructed by utilizing the undersampling model generated by the undersampling model generation method;
and reconstructing the image to be reconstructed through a neural network corresponding to the undersampling model to obtain a reconstructed image.
An undersampled model generation apparatus, the apparatus comprising:
the sampling module is used for obtaining an undersampled image according to the undersampled model to be updated;
the reconstruction module is used for reconstructing the undersampled image through a preset neural network to obtain a reconstructed image;
the acquisition module is used for acquiring the coding line with the minimum difference value when the reconstructed image does not meet the preset requirement;
the generating module is used for updating the under-sampling model to be updated according to the coding line and generating a new under-sampling model;
and the iteration module is used for taking the new undersampled model as an undersampled model to be updated, returning to the step of obtaining an undersampled image according to the undersampled model to be updated until the reconstructed image meets the preset requirement, and taking the undersampled model corresponding to the reconstructed image meeting the requirement as the finally generated undersampled model.
A computer device comprising a memory and a processor, the memory storing a computer program, wherein the processor when executing the computer program implements the steps of the undersampled model generation method of any of the above.
A computer-readable storage medium, having stored thereon a computer program which, when being executed by a processor, carries out the steps of the undersampled model generation method of any of the preceding claims.
According to the under-sampling model method, the under-sampling model device, the computer equipment and the storage medium, the under-sampling image is obtained by using the under-sampling model to be updated, and the under-sampling image is reconstructed through the preset neural network to obtain the reconstructed image. And then, whether the reconstructed image meets the preset requirement is determined by evaluating the reconstructed image, so that whether the under-sampling model achieves the optimal under-sampling condition is judged. And when the reconstructed image does not meet the preset requirement, acquiring a coding line with the minimum difference value, updating the under-sampling model according to the coding line with the minimum difference value to generate a new under-sampling model, obtaining the under-sampling image again according to the new under-sampling model, and evaluating after reconstruction again, thereby ensuring that the under-sampling model can be adjusted according to the neural network and ensuring that the under-sampling model which is most matched with the neural network model is obtained.
Drawings
FIG. 1 is a diagram of an application environment of a method for generating an undersampled model in one embodiment;
FIG. 2 is a schematic flow chart diagram of a method for generating an undersampled model in one embodiment;
FIG. 3 is a diagram illustrating an example of an iterative generation of an undersampled model;
FIG. 4 is a diagram of an undersampled model, an undersampled image, and a reconstructed image in one embodiment;
FIG. 5 is a flowchart illustrating the step of obtaining the code line with the smallest difference according to an embodiment;
FIG. 6 is a block diagram showing the structure of an undersampling model generation apparatus according to an embodiment;
FIG. 7 is a diagram illustrating an internal structure of a computer device according to an embodiment.
Detailed Description
In order to make the objects, technical solutions and advantages of the present application more apparent, the present application is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present application and are not intended to limit the present application.
The undersampling model generation method provided by the application can be applied to the application environment shown in fig. 1. The application environment relates to the terminal 102 and may also relate to the server 104. The terminal 102 communicates with the server 104 through a network. After the terminal 102 receives a generation instruction of a user, the terminal 102 can respond to the generation instruction to independently realize the undersampling model generation method. The terminal 102 may send the generation instruction to the server 104, and the server 104 responds to the generation instruction to implement the undersampling model generation method. Specifically, the server 104 obtains an undersampled image according to the undersampled model to be updated. The server 104 reconstructs the undersampled image through a preset neural network to obtain a reconstructed image. When the reconstructed image does not meet the preset requirement, the server 104 acquires the encoding line with the minimum difference. The server 104 generates a new undersampling model from the code line with the smallest difference. The server 104 takes the new undersampled model as the undersampled model to be updated, and returns to the step of obtaining the undersampled image by using the undersampled model to be updated until the reconstructed image meets the preset requirement, and the undersampled model corresponding to the reconstructed image meeting the requirement is taken as the finally generated undersampled model. The terminal 102 may be, but not limited to, various personal computers, notebook computers, smart phones, tablet computers, and portable wearable devices, and the server 104 may be implemented by an independent server or a server cluster formed by a plurality of servers.
In one embodiment, as shown in fig. 2, there is provided an undersampling model generation method, which is described by taking the method as an example applied to the server in fig. 1, and includes the following steps:
and step S202, obtaining an undersampled image according to the undersampled model to be updated.
The undersampling refers to a sampling mode that the sampling frequency of data is lower than twice of the highest frequency of a signal, and the undersampling model is a template for undersampling. It can be understood that, during scanning, the magnetic resonance apparatus performs sampling according to a set undersampling model, that is, performs data acquisition according to the specification of the undersampling model, and fills the acquired data into the K space until the data filling of the K space is completed. The to-be-updated undersampled model is an undersampled model needing to be updated, and can be an initial undersampled model generated initially or an undersampled model which is newly generated according to the initial undersampled model and does not meet requirements, and the initial undersampled model can be generated in a random undersampled mode. It can be understood that the undersampled model can be used as the undersampled model to be updated as long as it is determined that the acquired reconstructed image does not meet the requirement after the undersampled image acquired by the undersampled model is reconstructed. The undersampled image is an image reconstructed by the undersampled model from the K space data acquired by the K space. The K-space is a frequency domain space, and the imaging data is arranged at a specific K-space position according to different spatial frequencies and finally transformed into an image. That is, an undersampled image may be understood as an image derived from K-space data acquired by an undersampled model. And when the under-sampling model acquires partial data from the K space, transforming the acquired partial data to an image domain through inverse Fourier transform to obtain a corresponding under-sampling image.
Specifically, after the server receives a generation instruction sent by the terminal, the server calls a corresponding to-be-updated undersampled model by responding to the received generation instruction, data acquisition is performed from a fully-acquired K space by using the called to-be-updated undersampled model, and the acquired data are subjected to inverse Fourier transform to obtain a corresponding undersampled image. It can be understood that the under-sampled model to be updated acts on the K space of the fully sampled image, that is, the under-sampled model to be updated and the K space are subjected to point multiplication to obtain the under-sampled K space. And performing inverse Fourier transform on the undersampled K space to obtain an undersampled image. For example, when a user needs to update an under-sampling model or needs to generate a new under-sampling model, a generation instruction is issued by operating the terminal, and the terminal sends the generation instruction to the server. The server calls the corresponding to-be-updated undersampled model according to the received generation instruction, acquires undersampled K space data from the fully-acquired K space by using the to-be-updated undersampled model, and converts the undersampled K space data to an image domain to acquire a corresponding undersampled image.
And S204, reconstructing the undersampled image through a preset neural network to obtain a reconstructed image.
The reconstruction refers to image reconstruction, and the image reconstruction refers to high-speed mathematical operation on imaging data. The operation of conventional image reconstruction mainly uses fast fourier transform. In this embodiment, a preset neural network is used to reconstruct the undersampled image, so as to obtain a corresponding reconstructed image. The neural network is completed according to the training of the sample image in advance and can be applied to a model for image reconstruction. Neural networks include, but are not limited to, full convolutional networks, combat-generated networks, U-net networks, and residual networks, among others.
Specifically, after the corresponding undersampled image is obtained through the undersampled model, a pre-trained neural network is called to perform image reconstruction on the undersampled image generated by the undersampled model, and a reconstructed image output by the neural network can be obtained. For example, the server inputs the undersampled image into the neural network by calling the neural network deployed in the local in advance, so as to perform a series of operations such as convolution, pooling and the like on the undersampled image by using the neural network, thereby obtaining a reconstructed image. The image reconstruction by using the neural network is faster and more accurate than the traditional image reconstruction operation method.
In step S206, when the reconstructed image does not satisfy the preset requirement, the encoding line with the minimum difference is obtained.
The encoding line refers to a phase encoding line in a K space, and the encoding line with the minimum difference value is the phase encoding line with the minimum difference value between the phase encoding line in the K space corresponding to the reconstructed image and the phase encoding line in the K space corresponding to the full sampling image. And if the reconstructed image does not meet the preset requirement, comparing the reconstructed image with the full sampling image, and determining that the difference between the reconstructed image and the full sampling image is too large, determining that the reconstructed image reconstructed according to the undersampled image does not meet the requirement. The too large difference can be understood as that, when the reconstructed image is compared with the full sampling image based on the preset comparison index, and the obtained comparison value is greater than the set threshold value, the too large difference between the reconstructed image and the full sampling image can be determined.
In one embodiment, before acquiring the encoding line with the minimum difference when the reconstructed image does not satisfy the preset requirement in step S206, a step of evaluating whether the reconstructed image satisfies the preset requirement is further included.
The step of evaluating whether the reconstructed image meets the preset requirements specifically comprises: comparing the reconstructed image with the full sampling image based on a preset comparison index to obtain a comparison value; and when the reconstructed image is determined to not meet the preset requirement according to the comparison value and the threshold value, acquiring the encoding line with the minimum difference value.
The predetermined comparison Index includes, but is not limited to, PSNR (Peak Signal to Noise Ratio) and SSIM (Structural Similarity Index). A threshold value is set by carrying out weighted comprehensive evaluation on the difference between indexes such as PSNR (Peak Signal to noise ratio), SSIM (structural similarity) and the like of the reconstructed image and the fully sampled image. And when the weighted comprehensive evaluation index of the multiple indexes is smaller than the threshold value, the reconstructed image is considered to meet the preset requirement, step S212 is executed, otherwise, the reconstructed image is determined not to meet the requirement, step S206 is executed, and the encoding line with the minimum difference value is obtained to update the undersampling model.
Specifically, for the encoding line with the smallest difference, after the reconstructed image is obtained, a preset full sampling image is obtained. And transforming the reconstructed image and the full sampling image to a K space, namely transforming the reconstructed image and the full sampling image from an image domain to the corresponding K space by utilizing fast Fourier transform to obtain the K space corresponding to the reconstructed image and the K space of the full sampling image. And comparing the value of each point on the phase coding line in the K space corresponding to the reconstructed image with the value of each point on the coding line corresponding to the K space of the full sampling image, and determining the difference value between each point on each corresponding phase coding line in the K space, wherein the difference value is a comparison value and is used for representing the difference between the two values. And for the difference value of each pixel point on each coding line, comprehensively considering indexes such as L1 or L2 norm, covariance and the like of each pixel on the coding line, comparing the weighted comprehensive values of the indexes on the coding lines, and selecting the coding line with the minimum comprehensive difference value. When comparing the difference values, the code lines already belonging to the undersampled model are excluded.
In step S208, a new undersampling model is generated according to the code line with the minimum difference.
Specifically, after the code line with the minimum difference is acquired, the code line is added to the under-sampling model to be updated to obtain a new under-sampling model, that is, the generated new under-sampling model.
And step S210, taking the new undersampled model as the undersampled model to be updated, returning to the step S202, and obtaining the undersampled image according to the undersampled model to be updated until the reconstructed image meets the preset requirement.
And step S212, taking the undersampled model corresponding to the reconstructed image meeting the requirements as the finally generated undersampled model.
Specifically, after a new under-sampling model is generated, it is necessary to determine whether an under-sampled image acquired by the new under-sampling model can meet the requirement for reconstruction. The new undersampled model is taken as the undersampled model to be updated, the step S202 is returned, and the undersampled image is obtained according to the undersampled model to be updated. And then, performing the subsequent steps in the same way, namely calling a neural network to reconstruct the undersampled image to obtain a reconstructed image. When the reconstructed image is evaluated to be determined to meet the preset requirement, the undersampled model corresponding to the reconstructed image meeting the preset requirement can be directly used as the finally generated undersampled model, namely, the generation of a new undersampled model is stopped. Otherwise, when the reconstructed image is evaluated and determined to also not meet the preset requirement, iteration is continued, namely, the encoding line with the minimum difference value is obtained again and added into the under-sampling model to be updated, and a new under-sampling model is obtained again. That is, the under-sampling model is continuously updated iteratively, and only when the obtained reconstructed image meets the preset requirement, the iteration can be stopped, and the finally generated under-sampling model is obtained.
For example, referring to fig. 3, in the first iteration, the undersampled model 1 is used as the undersampled model 1 to be updated, the undersampled model 1 to be updated is used to obtain the undersampled image 1, and the undersampled image 1 is reconstructed by the neural network to obtain the reconstructed image 1. And when the reconstructed image 1 meets the preset requirement, taking the under-sampling model 1 to be updated as the finally generated under-sampling model, namely taking the under-sampling model 1 as the finally generated under-sampling model. And when the reconstructed image 1 does not meet the preset requirement, comparing the encoding line of the K space of the reconstructed image 1 with the encoding line of the K space of the full sampling image to obtain the encoding line 1 with the minimum difference value. And adding the coding line 1 with the minimum difference value into the undersampling model 1 (the undersampling model 1 to be updated) to obtain an undersampling model 2. The second round of iteration, namely, the undersampled model 2 is taken as the undersampled model 2 to be updated. Similarly, an under-sampled image 2 is obtained by using the under-sampled model 2 to be updated, and the under-sampled image 2 is reconstructed by a neural network to obtain a reconstructed image 2. And when the reconstructed image 2 meets the preset requirement, taking the under-sampling model 2 to be updated as the finally generated under-sampling model, namely taking the under-sampling model 2 as the finally generated under-sampling model. And when the reconstructed image 2 does not meet the preset requirement, comparing the encoding line of the K space of the reconstructed image 2 with the encoding line of the K space of the full sampling image to obtain the encoding line 2 with the minimum difference value. Adding the encoding line 2 with the minimum difference value into the undersampling model 2 (the undersampling model 2 to be updated) to obtain an undersampling model 3, taking the undersampling model 3 as the undersampling model 3 to be updated, and iterating the steps of the undersampling model 1 to be updated and the undersampling model 2 to be updated by the undersampling model 3 to be updated until the finally generated undersampling model can be obtained. The same is true for other under-sampled models to be updated, which is not described in detail herein. Further, referring to fig. 4, the under-sampled model to be updated is updated continuously, and the number of encoding lines of the current under-sampled model to be updated is increased compared with the previous under-sampled model to be updated, so that the acquired under-sampled image is closer to the full-sampled image, and the reconstructed image reconstructed by the neural network is also closer to the full-sampled image.
According to the under-sampling model method, the under-sampling model to be updated is used for obtaining the under-sampling image, and the under-sampling image is reconstructed through the preset neural network to obtain the reconstructed image. And then, whether the reconstructed image meets the preset requirement is determined by evaluating the reconstructed image, so that whether the under-sampling model achieves the optimal under-sampling condition is judged. And when the reconstructed image does not meet the preset requirement, acquiring a coding line with the minimum difference, updating the undersampled model according to the coding line with the minimum difference to generate a new undersampled model, obtaining the undersampled image again according to the new undersampled model, and evaluating after reconstruction, so that the undersampled model can be adjusted according to the neural network, and the undersampled model which is most matched with the neural network model is ensured to be obtained.
In one embodiment, the obtaining of the undersampled image according to the undersampled model to be updated comprises the following steps:
and generating an initial sampling model, and taking the initial sampling model as an undersampling model to be updated. And obtaining an undersampled image based on the undersampled model to be updated and a preset fully sampled image.
Specifically, the initial sampling model is an undersampled model that is generated for the first time. It can be understood that, after the server receives the generation instruction sent by the terminal, an initial undersampling model is randomly generated in response to the generation instruction, and the initial undersampling model is taken as the undersampling model to be updated. Further, the under-sampled model to be updated acts on the K space of the fully sampled image, that is, the under-sampled model to be updated and the K space are subjected to point multiplication to obtain the under-sampled K space. And performing inverse Fourier transform on the undersampled K space to obtain an undersampled image. In this embodiment, the corresponding under-sampled image is obtained through the under-sampled model, so that it is convenient to subsequently evaluate whether the under-sampled model needs to be updated through the under-sampled image.
In one embodiment, as shown in fig. 5, when it is determined that the reconstructed image does not satisfy the preset requirement according to the comparison value and the threshold value, acquiring the encoding line with the minimum difference value includes the following steps:
and step S502, when the reconstructed image is determined to not meet the preset requirement according to the comparison value and the threshold value, determining the existing coding line of the undersampled model corresponding to the reconstructed image.
Specifically, when it is determined that the reconstructed image does not meet the preset requirement, the undersampled model corresponding to the reconstructed image is determined. For example, the reconstructed image 1 is obtained by reconstructing an undersampled image acquired by the undersampled model 1 to be updated (the undersampled model 1), and the undersampled model 1 to be updated (the undersampled model 1) is an undersampled model corresponding to the reconstructed image 1. And if the reconstructed image is a reconstructed image 2 and the reconstructed image 2 is obtained by reconstructing an undersampled image acquired by the undersampled model 2 to be updated (the undersampled model 2), the undersampled model 2 to be updated (the undersampled model 2) is an undersampled model corresponding to the reconstructed image 2. Further, after an undersampling model corresponding to the reconstructed image is determined, the existing coding lines in the undersampling model are obtained. That is, after an under-sampling model corresponding to a reconstructed image is obtained, an under-sampling image corresponding to the under-sampling model is obtained, and a non-zero coding line in a K space corresponding to the under-sampling image is an existing coding line of the under-sampling model.
Step S504, existing coding lines are removed from the reconstructed image and the preset full sampling image K space respectively, and residual coding lines are obtained.
Specifically, after determining the existing encoding lines of the under-sampling model corresponding to the reconstructed image, the existing encoding lines are removed from the K space of the reconstructed image and the K space of the full-sampling image. For example, if the existing encoding lines are encoding line 1 and encoding line 2, then the encoding line 1 and the encoding line 2 are removed from the K space of the reconstructed image and the K space of the full sampling image, and the remaining encoding lines are the remaining encoding lines.
Step S506, calculating a difference between the reconstructed image and the remaining encoding lines in the K space of the preset full-sampling image, and obtaining the encoding line with the minimum difference.
Specifically, the difference between the reconstructed image and the residual encoding line in the K space of the preset fully sampled image is calculated, that is, the K space of the reconstructed image and the K space of the fully sampled image are subtracted. That is, the difference between the encoding line on the K space of the reconstructed image and the corresponding encoding line on the K space of the full sampling image is calculated, thereby obtaining the encoding line with the minimum difference. The difference includes, but is not limited to, an LI norm difference, an L2 norm difference, a covariance value of the coding line, or a combination of the above differences (LI norm difference, L2 norm difference, covariance value), and the like. After the difference value calculation is carried out, the value with the minimum difference value is determined through comparison between the difference values, and therefore the coding line corresponding to the value with the minimum difference value is obtained, namely the coding line with the minimum difference value is obtained. In the embodiment, the coding line with the minimum calculation difference is removed, so that the coding line is prevented from being repeated, and a newly generated undersampling model is not changed.
In one embodiment of the present invention, neural Networks are a pair-forming network and a full convolution network, wherein, a Generative Adaptive Networks (GAN) is a deeply learned network model. The antagonistic neural network comprises two parts: a network and a discrimination network are generated, wherein the network is used to generate images such that the generated images can confuse the discrimination network. The discrimination network is used for judging whether the image is a real image or an image generated by the generation network, and accurate judgment is required to be given as much as possible. The full convolution neural network can learn the mapping relation between the graphs, and accept input images with any size and can also be used for image generation.
Specifically, the undersampled image is reconstructed through a resist generation network or a full convolution network, so as to obtain a reconstructed image. Taking the countermeasure generation network as an example, in this embodiment, the under-sampled image is input to the generation network of the trained countermeasure generation network, and the image is reconstructed by the generation network of the countermeasure generation network according to the under-sampled image, so as to obtain a reconstructed image. The neural network obtained through the antagonistic training is used for image reconstruction, so that a high-quality reconstructed image can be guaranteed. Or, directly inputting the undersampled images into a full convolution network, performing convolution operation on the undersampled images by using the full convolution network, and extracting corresponding characteristics so as to obtain reconstructed images according to the characteristics. In the embodiment, the neural network is used for image reconstruction, and compared with the traditional image reconstruction method, the method is not only fast, but also improves the accuracy.
In an embodiment, an image reconstruction method is further provided, which specifically includes: and sampling the target object by using the undersampling model generated in the undersampling model generation method to obtain an image to be reconstructed. And reconstructing the image to be reconstructed through the neural network corresponding to the undersampling model to obtain a reconstructed image.
Specifically, after a final undersampled model is generated according to a technical scheme in the undersampled model generation method, the undersampled model can be used for image reconstruction. For example, when a magnetic resonance scanner is used to scan images of different sequence parameters of a target object, the undersampled model generated by the undersampled model generation method can be used to undersampled the target object, so as to acquire corresponding undersampled K-space data, the acquired undersampled K-space data is transformed from a K-space domain to an image domain through inverse fourier transform, and an undersampled image corresponding to the undersampled K-space data is acquired, where the undersampled image is an image to be reconstructed that needs to be subjected to image reconstruction. And then, calling the neural network, inputting the image to be reconstructed into the called neural network, and carrying out image reconstruction on the image to be reconstructed through the neural network to obtain a reconstructed image. In order to obtain a high-quality reconstructed image, the called neural network should be the neural network corresponding to the undersampling model. The neural network corresponding to the undersampled model can be understood as the neural network used in the undersampled model generation method. The finally generated undersampled model can be understood as the undersampled model obtained after the iteration of the undersampled model generation method is stopped, that is, if the reconstructed image 1 of the undersampled model 1 to be updated meets the requirement, the undersampled model used in the embodiment is the undersampled model 1 to be updated. If the reconstructed image 2 of the under-sampling model 2 to be updated meets the requirement, the under-sampling model used in the embodiment is the under-sampling model 2 to be updated, that is, the under-sampling model used in the last iteration round can be understood as the under-sampling model. For example, it is assumed that the called neural network is a countermeasure type generation network, that is, after an undersampled model matched with the countermeasure type generation network is obtained according to the undersampled model generation method, the undersampled model is used to undersample the target object, and an undersampled image corresponding to the undersampled model is obtained, and the undersampled image is the image to be reconstructed that needs to be reconstructed. Then, a neural network matched with the undersampled model is called, namely, a confrontational generation network is called. And inputting the image to be reconstructed into the countermeasure type generation network, and carrying out image reconstruction on the image to be reconstructed by utilizing the generation network in the countermeasure type generation network to obtain a reconstructed image.
In addition, it should be understood that both the neural network in the image reconstruction method of the present embodiment and the neural network in the under-sampling model generation method are network models that are trained in advance by the training image and the target image and can be used for image reconstruction. The training image is an undersampled image sample and is used as input of neural network training, and the target image is a fully sampled image sample and is used as a target of the neural network training.
In the embodiment, the undersampled model is used for sampling during magnetic resonance scanning, so that the scanning time is shortened, the undersampled model matched with the neural network can be generated by the undersampled model generation method, and after the undersampled model generated by the undersampled model generation method is used for undersampled the target object to obtain the undersampled image, the undersampled image is reconstructed according to the matched neural network, so that the reconstructed image is closer to the fully-acquired image, and the accuracy of image reconstruction is improved.
It should be understood that although the steps in the flowcharts of fig. 2 and 5 are shown in sequence as indicated by the arrows, the steps are not necessarily performed in sequence as indicated by the arrows. The steps are not performed in the exact order shown and described, and may be performed in other orders, unless explicitly stated otherwise. Moreover, at least some of the steps in fig. 2 and 5 may include multiple sub-steps or multiple stages, which are not necessarily performed at the same time, but may be performed at different times, and the order of performing the sub-steps or stages is not necessarily sequential, but may be performed alternately or alternately with other steps or at least some of the sub-steps or stages of other steps.
In one embodiment, as shown in fig. 6, there is provided an undersampling model generation apparatus including: a sampling module 602, a reconstruction module 604, an acquisition module 606, a generation module 608, and an iteration module 610, wherein:
the sampling module 602 is configured to obtain an undersampled image according to the undersampled model to be updated.
The reconstructing module 604 is configured to reconstruct the undersampled image through a preset neural network to obtain a reconstructed image.
An obtaining module 606, configured to obtain, when the reconstructed image does not meet the preset requirement, the encoding line with the smallest difference.
The generating module 608 is configured to update the under-sampling model to be updated according to the encoding line, and generate a new under-sampling model.
And the iteration module 610 is configured to use the new undersampled model as the undersampled model to be updated, return to the step of obtaining the undersampled image according to the undersampled model to be updated until it is determined that the reconstructed image meets the preset requirement, and use the undersampled model corresponding to the reconstructed image meeting the requirement as the finally generated undersampled model.
In one embodiment, the sampling module 602 is further configured to generate an initial sampling model, and use the initial sampling model as an under-sampling model to be updated; and obtaining an undersampled image based on the undersampled model to be updated and a preset fully sampled image.
In one embodiment, the obtaining module 606 is further configured to compare the reconstructed image with the full sampling image based on a preset comparison index to obtain a comparison value; and when the reconstructed image is determined to not meet the preset requirement according to the comparison value and the threshold value, acquiring the encoding line with the minimum difference value.
In one embodiment, the obtaining module 606 is further configured to determine an existing encoding line of the under-sampled model corresponding to the reconstructed image when it is determined that the reconstructed image does not meet the preset requirement according to the comparison value and the threshold; respectively removing the existing coding lines from the reconstructed image and the preset full sampling image K space to obtain residual coding lines; and calculating the difference value of the residual coding lines on the reconstructed image and the preset full sampling image K space to obtain the coding line with the minimum difference value.
In one embodiment, the generating module 608 is further configured to add the encoding line to the under-sampled model to be updated, and generate a new under-sampled model.
For specific limitations of the undersampled model generation apparatus, reference may be made to the above limitations of the undersampled model generation method, which are not described herein again. The modules in the undersampling model generation device can be wholly or partially realized by software, hardware and a combination thereof. The modules can be embedded in a hardware form or independent from a processor in the computer device, and can also be stored in a memory in the computer device in a software form, so that the processor can call and execute operations corresponding to the modules.
In one embodiment, a computer device is provided, which may be a server, and the internal structure thereof may be as shown in fig. 7. The computer device includes a processor, a memory, a network interface, and a database connected by a system bus. Wherein the processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device comprises a nonvolatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operating system and the computer program to run on the non-volatile storage medium. The database of the computer device is used for storing data. The network interface of the computer device is used for communicating with an external terminal through a network connection. The computer program is executed by a processor to implement a method of undersampling model generation.
Those skilled in the art will appreciate that the architecture shown in fig. 7 is merely a block diagram of some of the structures associated with the disclosed aspects and is not intended to limit the computing devices to which the disclosed aspects apply, as particular computing devices may include more or less components than those shown, or may combine certain components, or have a different arrangement of components.
In one embodiment, a computer device is provided, comprising a memory and a processor, the memory having a computer program stored therein, the processor implementing the following steps when executing the computer program:
obtaining an undersampled image according to the undersampled model to be updated;
reconstructing the undersampled image through a preset neural network to obtain a reconstructed image;
when the reconstructed image does not meet the preset requirement, acquiring a coding line with the minimum difference value;
updating the under-sampling model to be updated according to the coding line to generate a new under-sampling model;
and taking the new undersampled model as an undersampled model to be updated, returning to the step of obtaining an undersampled image according to the undersampled model to be updated until the reconstructed image meets the preset requirement, and taking the undersampled model corresponding to the reconstructed image meeting the requirement as the finally generated undersampled model.
In one embodiment, the processor, when executing the computer program, further performs the steps of:
generating an initial sampling model, and taking the initial sampling model as an undersampling model to be updated; and obtaining an undersampled image based on the undersampled model to be updated and a preset fully sampled image.
In one embodiment, the processor, when executing the computer program, further performs the steps of:
comparing the reconstructed image with the full sampling image based on a preset comparison index to obtain a comparison value; and when the reconstructed image is determined to not meet the preset requirement according to the comparison value and the threshold, acquiring the encoding line with the minimum difference value.
In one embodiment, the processor, when executing the computer program, further performs the steps of:
when the reconstructed image is determined to not meet the preset requirement according to the comparison value and the threshold value, determining the existing coding line of the undersampled model corresponding to the reconstructed image; respectively removing the existing coding lines from the reconstructed image and the preset full sampling image K space to obtain residual coding lines; and calculating the difference value of the residual coding lines on the reconstructed image and the preset full sampling image K space to obtain the coding line with the minimum difference value.
In one embodiment, the processor, when executing the computer program, further performs the steps of:
and adding the coding line into the undersampling model to be updated to generate a new undersampling model.
In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored which, when executed by a processor, performs the steps of:
obtaining an undersampled image according to the undersampled model to be updated;
reconstructing the undersampled image through a preset neural network to obtain a reconstructed image;
when the reconstructed image does not meet the preset requirement, acquiring a coding line with the minimum difference value;
updating the under-sampling model to be updated according to the coding line to generate a new under-sampling model;
and taking the new undersampled model as an undersampled model to be updated, returning to the step of obtaining an undersampled image according to the undersampled model to be updated until the reconstructed image meets the preset requirement, and taking the undersampled model corresponding to the reconstructed image meeting the requirement as the finally generated undersampled model.
In one embodiment, the computer program when executed by the processor further performs the steps of:
generating an initial sampling model, and taking the initial sampling model as an undersampling model to be updated; and obtaining an undersampled image based on the undersampled model to be updated and a preset fully sampled image.
In one embodiment, the computer program when executed by the processor further performs the steps of:
comparing the reconstructed image with the full sampling image based on a preset comparison index to obtain a comparison value; and when the reconstructed image is determined to not meet the preset requirement according to the comparison value and the threshold value, acquiring the encoding line with the minimum difference value.
In one embodiment, the computer program when executed by the processor further performs the steps of:
when the reconstructed image is determined to not meet the preset requirement according to the comparison value and the threshold value, determining the existing coding line of the undersampled model corresponding to the reconstructed image; respectively removing the existing coding lines from the reconstructed image and the preset full sampling image K space to obtain residual coding lines; and calculating the difference value of the residual coding lines on the reconstructed image and the preset full sampling image K space to obtain the coding line with the minimum difference value.
In one embodiment, the computer program when executed by the processor further performs the steps of:
and adding the coding line into the undersampled model to be updated to generate a new undersampled model.
It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above can be implemented by hardware instructions of a computer program, which can be stored in a non-volatile computer-readable storage medium, and when executed, can include the processes of the embodiments of the methods described above. Any reference to memory, storage, database, or other medium used in the embodiments provided herein may include non-volatile and/or volatile memory, among others. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically Programmable ROM (EPROM), electrically Erasable Programmable ROM (EEPROM), or flash memory. Volatile memory can include Random Access Memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms such as Static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double Data Rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous Link DRAM (SLDRAM), rambus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).
The technical features of the above embodiments can be arbitrarily combined, and for the sake of brevity, all possible combinations of the technical features in the above embodiments are not described, but should be considered as the scope of the present specification as long as there is no contradiction between the combinations of the technical features.
The above-mentioned embodiments only express several embodiments of the present application, and the description thereof is specific and detailed, but not to be understood as limiting the scope of the invention. It should be noted that, for a person skilled in the art, several variations and modifications can be made without departing from the concept of the present application, and these are all within the scope of protection of the present application. Therefore, the protection scope of the present patent application shall be subject to the appended claims.
Claims (10)
1. A method of generating an undersampled model, the method comprising:
obtaining an undersampled image according to an undersampled model to be updated;
reconstructing the undersampled image through a preset neural network to obtain a reconstructed image;
when the reconstructed image does not meet the preset requirement, acquiring a coding line with the minimum difference value;
generating a new undersampling model according to the coding line with the minimum difference value;
and taking the new undersampled model as an undersampled model to be updated, returning to the step of obtaining an undersampled image according to the undersampled model to be updated until the reconstructed image meets the preset requirement, and taking the undersampled model corresponding to the reconstructed image meeting the requirement as the finally generated undersampled model.
2. The method of claim 1, wherein the step of obtaining an undersampled image from an undersampled model to be updated comprises:
generating an initial sampling model, and taking the initial sampling model as an undersampling model to be updated;
and obtaining an undersampled image based on the undersampled model to be updated and a preset fully sampled image.
3. The method according to claim 2, wherein the step of obtaining the encoding line with the minimum difference value when the reconstructed image does not satisfy the preset requirement comprises:
comparing the reconstructed image with the fully sampled image based on a preset comparison index to obtain a comparison value;
and when the reconstructed image is determined to not meet the preset requirement according to the comparison value and the threshold value, acquiring the encoding line with the minimum difference value.
4. The method according to claim 3, wherein when it is determined that the reconstructed image does not satisfy the preset requirement according to the comparison value and the threshold, the step of obtaining the encoding line with the smallest difference comprises:
when the reconstructed image is determined to not meet the preset requirement according to the comparison value and the threshold value, determining the existing coding line of the undersampled model corresponding to the reconstructed image;
respectively removing the existing coding lines from the reconstructed image and a preset full sampling image K space to obtain residual coding lines;
and calculating the difference value of the residual coding lines on the reconstructed image and a preset full sampling image K space to obtain the coding line with the minimum difference value.
5. The method of claim 1, wherein generating a new undersampled model from the encoded line with the smallest difference comprises:
and adding the coding line into the undersampled model to be updated to generate a new undersampled model.
6. The method of claim 1, wherein the neural network is a challenge generation network or a full convolution network.
7. A method of image reconstruction, the method comprising:
sampling a target object by using an undersampling model generated by the undersampling model generating method of any one of claims 1 to 6 to obtain an image to be reconstructed;
and reconstructing the image to be reconstructed through a neural network corresponding to the undersampling model to obtain a reconstructed image.
8. An undersampled model generation apparatus, the apparatus comprising:
the sampling module is used for obtaining an undersampled image according to the undersampled model to be updated;
the reconstruction module is used for reconstructing the undersampled image through a preset neural network to obtain a reconstructed image;
the acquisition module is used for acquiring the coding line with the minimum difference value when the reconstructed image does not meet the preset requirement;
the generating module is used for updating the under-sampling model to be updated according to the coding line and generating a new under-sampling model;
and the iteration module is used for taking the new undersampled model as an undersampled model to be updated, returning to the step of obtaining an undersampled image according to the undersampled model to be updated until the reconstructed image meets the preset requirement, and taking the undersampled model corresponding to the reconstructed image meeting the requirement as the finally generated undersampled model.
9. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor, when executing the computer program, implements the steps of the method of any of claims 1 to 7.
10. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the steps of the method of any one of claims 1 to 7.
Priority Applications (3)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910789671.3A CN110570487B (en) | 2019-08-26 | 2019-08-26 | Undersampling model generation method, image reconstruction method, device and computer equipment |
| US17/002,817 US11568584B2 (en) | 2019-08-26 | 2020-08-26 | Systems and methods for magnetic resonance imaging |
| US18/157,794 US11880915B2 (en) | 2019-08-26 | 2023-01-20 | Systems and methods for magnetic resonance imaging |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910789671.3A CN110570487B (en) | 2019-08-26 | 2019-08-26 | Undersampling model generation method, image reconstruction method, device and computer equipment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110570487A CN110570487A (en) | 2019-12-13 |
| CN110570487B true CN110570487B (en) | 2022-11-25 |
Family
ID=68776238
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910789671.3A Active CN110570487B (en) | 2019-08-26 | 2019-08-26 | Undersampling model generation method, image reconstruction method, device and computer equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110570487B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111157935B (en) * | 2019-12-31 | 2022-04-26 | 上海联影智能医疗科技有限公司 | Magnetic resonance imaging method, magnetic resonance imaging device, storage medium and computer equipment |
| CN113359076B (en) * | 2020-03-06 | 2022-09-27 | 上海联影医疗科技股份有限公司 | Magnetic resonance imaging method, magnetic resonance imaging apparatus, computer device, and storage medium |
| CN111681297B (en) * | 2020-05-28 | 2024-02-27 | 上海联影智能医疗科技有限公司 | Image reconstruction method, computer device, and storage medium |
Family Cites Families (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103384836B (en) * | 2010-12-22 | 2016-10-26 | 皇家飞利浦电子股份有限公司 | Method and apparatus for nuclear magnetic resonance |
| RU2672151C2 (en) * | 2014-03-28 | 2018-11-12 | Конинклейке Филипс Н.В. | Correction of epi extraneous echoes |
| CN106597333B (en) * | 2016-12-30 | 2019-05-31 | 上海联影医疗科技有限公司 | A kind of magnetic resonance parallel imaging method and magnetic resonance imaging system |
| CN107576925B (en) * | 2017-08-07 | 2020-01-03 | 上海东软医疗科技有限公司 | Magnetic resonance multi-contrast image reconstruction method and device |
| US11029381B2 (en) * | 2018-01-12 | 2021-06-08 | Korea Advanced Institute Of Science And Technology | Method for varying undersampling dimension for accelerating multiple-acquisition magnetic resonance imaging and device for the same |
| CN109325985B (en) * | 2018-09-18 | 2020-07-21 | 上海联影智能医疗科技有限公司 | Magnetic resonance image reconstruction method, apparatus and computer readable storage medium |
-
2019
- 2019-08-26 CN CN201910789671.3A patent/CN110570487B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN110570487A (en) | 2019-12-13 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110570487B (en) | Undersampling model generation method, image reconstruction method, device and computer equipment | |
| Chantas et al. | Bayesian restoration using a new nonstationary edge-preserving image prior | |
| CN110827369B (en) | Undersampling model generation method, image reconstruction method, apparatus and storage medium | |
| CN110555526B (en) | Neural network model training method, image recognition method and device | |
| CN111383741B (en) | Method, device and equipment for establishing medical imaging model and storage medium | |
| CN112784981A (en) | Training sample set generation method, and training method and device for deep generation model | |
| CN111028153A (en) | Image processing and neural network training method and device and computer equipment | |
| CN113807353A (en) | Image conversion model training method, device, equipment and storage medium | |
| CN111157935B (en) | Magnetic resonance imaging method, magnetic resonance imaging device, storage medium and computer equipment | |
| CN111383742A (en) | Method, device, equipment and storage medium for establishing medical imaging model | |
| Vasudeva et al. | Co-VeGAN: Complex-valued generative adversarial network for compressive sensing MR image reconstruction | |
| CN115147402A (en) | Self-contrast learning for image processing | |
| CN111105471B (en) | Method and apparatus for generating sampling mask related to imaging | |
| CN111714124A (en) | Magnetic resonance film imaging method, device, imaging equipment and storage medium | |
| US20220366539A1 (en) | Image processing method and apparatus based on machine learning | |
| CN112001839B (en) | Cross-domain image conversion method based on semantic feature transformation, computer device and storage medium | |
| CN111199513B (en) | Image processing method, computer device, and storage medium | |
| CN112509120A (en) | System and method for reconstructing an image | |
| CN111161239B (en) | Medical image analysis method, device, storage medium and computer equipment | |
| US20240062438A1 (en) | Systems and methods for processing medical images with invertible neural networks | |
| CN111899163B (en) | Efficient structure preservation to generate single image super-resolution in an antagonistic network | |
| CN114494876A (en) | Image down-sampling method and image down-sampling system | |
| KR20240006470A (en) | Method, program, and apparatus for improving quality of medical data based on deep learning | |
| CN112802148A (en) | Image reconstruction method, target object detection method, apparatus, device, and medium | |
| CN117726542A (en) | Controllable noise removing method and system based on diffusion model |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |